Home

AI, Never Forget. — An independent observatory of AI memory
AI, never forget.
An independent observatory of AI memory

AI, Never forget

Not a slogan. An instruction.

We track, test and explain how artificial intelligence remembers, forgets and relearns — across models, updates and time.

Explore the Lab View the Memory Index

Every AI system remembers something. Almost none remember well.

Ask an AI to recall something you mentioned an hour ago, and the answer depends on more than the model itself — on how long the conversation has run, how the system was built to hold context, and what it was designed to let go of.

Some information survives. Most quietly doesn't. Context gets compressed, recall gets approximate, and what feels like memory is often something closer to a good guess.

AINeverForget exists to track exactly where that line falls — model by model, update by update.

The Lab

Original experiments on memory, context, recall and degradation.

Run on real models, published with full method — including where the method falls short.

First experiment

5 AIs. One Memory.

In preparation
One memory
↓
AI 01 · AI 02 · AI 03 AI 04 · AI 05
↓
Recall test

The same set of facts. The same conversation length. Five different models. How much actually survives — and how does it fail?

The Memory Index

A living, comparative record — across dimensions, not one score.

Memory isn't one number. It's tracked here across the ways it actually breaks down.

Context retention How much earlier information survives as context grows.
Recall Whether previously supplied information can be recovered accurately.
Cross-session memory Whether information persists once a conversation ends and a new one begins.
Instruction retention Whether earlier instructions are still followed as a conversation continues.
Contradiction handling How a model responds when new information conflicts with what it was told before.
Retrieval How reliably relevant information surfaces when it's needed, not just stored.
Long-context degradation How accuracy changes as conversations grow longer.
Baseline begins with Experiment 001.
The Record

Explanations, history and analysis — the reading behind the research.

Context

What "context window" actually means

The jargon behind why AI forgets mid-conversation, explained without the hand-waving.

Memory

A short history of AI memory

From stateless chatbots to systems that carry something forward — how we got here.

Knowledge

Building a knowledge system an AI can actually use

Personal knowledge management is quietly becoming an AI memory problem.

Why this exists

Memory is becoming part of intelligence. Someone should keep track.

AINeverForget observes, tests and records how artificial memory evolves — across models, across updates, over time.

It's built around one question the rest of the industry mostly explains and rarely tests: what does AI actually remember? The record, the research and the index are how we find out.